Recent news of Perplexity’s advertising efforts running aground and persistent rumors of OpenAI’s plans to speed onto the advertising highway have spurred me out of my writing slumber.
Many tokens have been spent hypothesizing ad formats and approaches these AI platforms might take to monetize their free userbase, but I’ve been surprised to see little commentary there has been on the opportunities these platforms have to re-invent advertising business models altogether.
The Opportunity
It has been just over 30 years since the banner ad was first served on the internet, 25 years since Google launched its search ads business, about 20 years since Meta launched its social ads business, and a little over 10 years since Amazon launched its sponsored products business.
Each of these offerings was revolutionary in that they (1) were perfectly adapted to the demand the platform was aggregating (Google: information, Facebook: social connections, Amazon: ecommerce), (2) offered a uniquely native mechanism for advertisers to communicate with audiences, and (3) subsidized the online service in a way that compounded the platform’s growth.
AI platforms now stand at the precipice of a similar opportunity – if they can make it work. But because advertising and commerce have always evolved hand in hand, getting there will require rethinking some of the first principles of both.
Theory of Advertising
Economics tells us that advertising creates consumer value in two ways. The first is informational, in that it reduces the costs borne by consumers in identifying products that meet their needs. The second is persuasive, in that it modifies a consumer’s utility function to spur demand for products (for example, by inspiring brand loyalty or aspiration).
As I revisit these old ideas, I observe that (1) advertising only exists in the context of the sale of a product or service, and (2) advertising might only create value if information asymmetries exist.
Agentic Future
AI platforms and agents today can already scour and synthesize the web autonomously for products and services that meet specified criteria. Thanks to the last few decades of work by Google and Amazon, the world now has a reasonably broad – if imperfect – global database of products and services (we can argue about its coarse vs. fine-grained nature, but that’s beside the point).
In other words, there’s been good progress in automating the merchant’s side of the marketplace (i.e. bringing SKUs to the internet), but the customer’s experience of the marketplace has remained largely static. Users still expend costs to search through a vast product space to identify their ideal purchase. These non-zero search costs may also be what continues to support the need and value of advertising online.
However, agentic commerce – the use of AI agents to complete transactions on a customer’s behalf – is a new paradigm awaiting AI platforms. If agents ultimately end up bearing the costs of search, it stands to reason that the informational value of advertising should fall. If agents begin to hold the memories, aspirations, and context of customers, and act more as trusted advisors less swayed by traditional branding, then perhaps the persuasive value of advertising falls too.
So are ads doomed? I don’t think we’re condemned to a monochromatic world of hyper-rationality, but what I see a broader reckoning on the horizon and a squeezing out of the “messy middle” that has long mediated the relationship between sellers and buyers on the internet. I don’t think we know yet how this will shake out, but its an area I aim to spend more time thinking about.
Further Reading
I enjoyed this discussion on the Rise of Agentic Commerce with Emily Glassberg Sands (Head of Information at Stripe). I also enjoyed reviewing the sample code included in Google’s newly announced Agent Payments Protocol (AP2) specifications.
Dispatches from the Frontier 